JP2002328920A - Method for associating words in parallel translation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for associating words in parallel translation with high rate of reappearance and high precision. SOLUTION: With respect to a method for associating words in parallel translation consisting of a sentence in the original language and a corresponding sentence in the translation language, the present invention consists of the first step which computes a first word correspondence rate for each word to word correspondence between the sentence in the original language and the sentence in the translation language utilizing a bilingual dictionary for the two languages and the second step which finds a combination of word correspondence having maximum sum of the first word correspondence rate in the parallel translation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a method for associating a bilingual sentence with a word.

[0002]

2. Description of the Related Art At present, developments are being made to improve the accuracy of a machine translation system, and research is being actively conducted on automatically creating a translation dictionary from a bilingual sentence in a target field.

[0003] Here, we focus on the word correspondence of the bilingual sentence, which is the starting point of the automation of the construction of the machine translation system.
The correspondence between words in the bilingual sentence is to determine the translation of each word in the bilingual sentence between the source language and the target language. In this work, the previously proposed methods (see references 1 to 7) have relied mainly on statistical information, but unfortunately there is not enough evidence for infrequent words. , The proportion (recall) of the associated words was low.

Reference 1: Gale, WA and Church KW (19
91) A program for aligning sentences in bilingual
corpora. Proc. of 29th ACL, pp. 177-184. Reference 2: Brown, PF, Della Pietra, SA, Della Pi
etra, VJ and Merer, RL (1993) The mathematics o
f statistical machine Translation: Parameter Estim
ation, Computational Linguistics, 19, 2, pp. 263-3
111. Reference 3: Dagan, I., Church, KW, and Gale, WA (1
994) Robust bilingual word alignment for Machine-A
ided Translation, Proc.of 4th ANLP, pp.34-40.

Reference 4: Wu, D. (1994) Aligning a parall
el English-Chinese corpus statistically with lexic
al criteria, Proc. of 32th ACL, pp.80-87. Reference 5: Smadja, F., MaKeown, K., and Hatzsivassilo
glou (1996) Translating collocations for bilingual
lexicons: A statistical approach, Computational L
inguistics, 21 (4), pp.1-38. Reference 6: Tanaka, T. and Matsuo, Y. (1999) Extractin
g Translation Equivalents from Non-parallel Corpor
a, Proc. of 8th TMI, pp.88-97. Reference 7: Sumita, E. (2000) Word alignment using matr
ix, Proc. of 6th PRICAI, pp.821.

[0006]

SUMMARY OF THE INVENTION An object of the present invention is to provide a method for associating a bilingual sentence with a high recall and high accuracy.

[0007]

According to a first aspect of the present invention, there is provided a method for associating a bilingual sentence with a source language sentence and a corresponding target language sentence with a word between the source language sentence and the target language sentence. In a first step of calculating a first word correspondence degree using a bilingual dictionary between both languages for each word correspondence of a word pair and one word, and in a bilingual sentence, the sum of the first word correspondence degrees is Second to find the maximum word correspondence combination
It is characterized by having steps.

According to a second aspect of the present invention, in the first aspect of the present invention, the first word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is as follows:
The bilingual dictionary is characterized in that it is determined based on whether or not it matches any of the translations of the words in the source language.

According to a third aspect of the present invention, in the first and second aspects of the present invention, a part of the entire bilingual sentence that has not yet been correlated in the second step is one word to one word between the two languages. A third step of calculating a second degree of word correspondence by using a thesaurus for each of the two languages for each word correspondence, and a portion of the entire bilingual sentence that has not yet been associated by the second step And
The method is characterized by including a fourth step of finding a combination of word correspondences that maximizes the second word correspondence degree.

According to a fourth aspect of the present invention, in the third aspect of the invention, the second word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is obtained from the source language obtained by a thesaurus. Are calculated based on a set of semantic classes to which the word belongs and a set of semantic classes to which the word of the target language belongs.

According to a fifth aspect of the present invention, in the invention of the fourth aspect, the second word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is obtained by using the source language obtained by a thesaurus. And the ratio of the size of the common set to the size of the union of the set of semantic classes to which the word belongs and the set of semantic classes to which the words of the target language belong.

According to a sixth aspect of the present invention, in the third to fifth aspects of the present invention, a portion of the entire bilingual sentence that has not been matched by the second step and the fourth step is set to one language between the two languages. For each word-to-word correspondence,
Using a part-of-speech table that represents the correspondence between parts of speech in both languages,
The fifth step of calculating the third word correspondence degree, and the sum of the third word correspondence degrees is the maximum in the part of the whole bilingual sentence that has not yet been associated by the second step and the fourth step. And a sixth step of obtaining a combination of word correspondences.

According to a seventh aspect of the present invention, in the invention of the sixth aspect, the third word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is such that the part of speech of the word in the target language is In the part-of-speech table, the determination is made based on whether or not the part-of-speech corresponding to the part of speech of the source language word matches.

According to an eighth aspect of the present invention, in the sixth to seventh aspects, a bilingual sentence partially corresponding in the second, fourth and sixth steps uses a compound word dictionary between two languages. A seventh step of associating one word-to-synthetic word or a compound word-to-synthetic word.

According to a ninth aspect of the present invention, the bilingual sentence partially corresponded in the second step, the fourth step, the sixth step, and the seventh step in the translation of the composite word between the two languages. The present invention is characterized in that an eighth step of associating one word-to-synthesis word or a synthetic word-to-synthesis word using a synthetic word part-of-speech table representing the correspondence between parts of speech is provided.

According to a tenth aspect of the present invention, in the ninth aspect, it is assumed that the word order of the source language sentence and the target language sentence constituting the bilingual sentence is substantially the same, and In a portion that has not yet been associated in the fourth, sixth, seventh, and eighth steps, a ninth step is provided for associating positional correspondences with each other.

According to an eleventh aspect of the present invention, in the first and second aspects of the present invention, in the bilingual sentence partially corresponded in the second step, a one word pair compound word or A tenth step of associating a compound word with a compound word is provided.

According to a twelfth aspect of the present invention, in the eleventh aspect, assuming that the word order of the source language sentence and the target language sentence constituting the bilingual sentence is substantially the same, the second step and the An eleventh step of associating a positionally corresponding part with a part which has not yet been associated in the tenth step is characterized in that:

According to a thirteenth aspect of the present invention, in the invention of the eleventh aspect, in the bilingual sentence partially corresponded in the second step and the tenth step, a compound word part of speech indicating a correspondence of a part of speech of a compound word between two languages. A first method for associating one word to a compound word or a compound word to a compound word using a table
It is characterized by having two steps.

According to a fourteenth aspect of the present invention, assuming that the word order of the source language sentence and the target language sentence constituting the bilingual sentence is substantially the same, the second step of In a portion which has not yet been associated by the tenth step and the twelfth step, a thirteenth step of associating positional correspondences is provided.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS [1] Explanation of the basic concept of a method of associating a bilingual sentence with a word according to the present invention

Recently, in order to overcome the problem of low recall, Document 8 proposes to use a bilingual thesaurus. Document 9 proposes to use similarities that exist between a dictionary and two languages from several linguistic viewpoints.

Reference 8: Ker, J., S. and Chang, SJ
(1997) A Class-Based Approach ToWord Alignment, Co
Mputational linguistics, 23, 2 pp. 313-343. Reference 9: Huang, J. and Choi, K. (2000) Chinese-Kor
ean Word-Alignment Based on Linguistic Comparison,
Proc. Of 38th ACL, pp. 392-399.

We utilize vocabulary knowledge of words, semantic classes and part of speech (POS) along this line. First, we use a bilingual dictionary between the source language and the target language, a synonym dictionary for each language, and a part-of-speech table indicating the correspondence between the parts of speech between the two languages. A score representing the degree of word correspondence for each word is calculated. Then, using dynamic programming, a search is made for a word-correspondence combination that maximizes the sum of scores in the bilingual sentence.

Next, based on a compound word dictionary between two languages and a compound word part-of-speech table showing the correspondence between the parts of speech of the compound words between the two languages, a compound word (idiom, phrase) is found to find the word correspondence. Fix it. As the bilingual dictionary, thesaurus, the compound word dictionary, the part-of-speech table, and the compound part-of-speech table, computerized (database) parts are used.

Our method is suitable for bilingual sentences that are considered to be similar to each other, for example, bilingual sentences between Japanese and Korean, but are suitable for Korean, Japanese, Uruguay, Turkish Bilingual translations of languages closely related to other languages, not limited to Altai languages such as Arabic, Mongolian, etc. It can also handle bilingual sentences in Italic languages such as words.

[2] Description of Embodiment of the Invention [2-1] Description of Overall Processing Procedure

FIG. 1 shows an overall processing procedure of a method of associating a bilingual sentence with a word.

Here, a method for associating a bilingual sentence with words will be described, taking as an example a case where the source language is Korean and the target language is Japanese.

The source language sentence in the bilingual sentence is divided into words, and each word is given a part of speech (step 1).
Similarly, the sentence of the target source language in the bilingual sentence is also divided into words, and each word is given a part of speech (step 2).

Next, using a bilingual dictionary between the source language and the target language, a synonym dictionary for each of the two languages, and a part-of-speech table representing the correspondence between the parts of speech between the two languages, one-to-one word correspondence. Processing is performed (step 3).

Thereafter, the process of associating the synthesized words is performed using the synthesized word dictionary between the two languages and the synthesized word part-of-speech table indicating the correspondence between the parts of speech of the synthesized words between the two languages (step 4). Finally, an associating process is performed on the uncorresponding word (step 5).

[2-2] Description of Word-to-One-Word Word Associating Process The score used in the one-word to one-word word associating process will be described. There are three types of scores: WORD_SCORE (first word correspondence), SEM_SCORE (second word correspondence), and POS_SCORE (third word correspondence).

(I) WORD_SCORE A bilingual dictionary between a source language and a target language can be regarded as a set of pairs of a source language word and a list of possible translations thereof. In the bilingual dictionary, if the word (target word) j of the target language is equal to one of the translations of the word (source word) k of the source language, the word between the source word k and the target word j is determined. WORD_SCOR
Assign 1 as the value of E (k, j). Otherwise, assign 0 as the value of WORD_SCORE.

(Ii) SEM_SCORE Here, it is assumed that the synonym dictionary of the source language and the synonym dictionary of the target language both have the same hierarchy, and that each word belongs to a semantic class according to its meaning. In other words, if a word is ambiguous, it belongs to more than one semantic class. A set of semantic classes to which the word x belongs is represented by SEM (x).

For the calculation of SEM_SCORE, for example, a Tanimoto coefficient (see Reference 10) often used when calculating similarity between documents in information retrieval can be used. Ta
The nimoto coefficient is also called the Jaccard coefficient. The Tanimoto coefficient is a ratio of the number of common elements (semantic classes) to the total number of elements (semantic classes), and takes a value between 0 and 1.

Reference 10: Kohonen, T. (1989) Self-organ
ization and Associative Memory, Springer-Verlag.

SEM_SCORE (k, j) between a source word k and a target word j is expressed by the following equation (1).
A set of semantic classes SEM (k) and SEM (j) common to the size of the union of the semantic class set SEM (k) of the source word k and the semantic class set SEM (j) of the target word j Expressed as a ratio of the size of the set.

[0039]

(Equation 1)

It should be noted that SEM_SCORE (k, j) may be calculated based on any of the following equations (2) to (5).

[0041]

(Equation 2)

(Iii) POS_SCORE First, a part-of-speech table indicating the correspondence of part-of-speech (POS: Part-Of-Speech) between the two languages will be described. This part-of-speech table is a table showing the correspondence between the part-of-speech of the source word and the part-of-speech of the possible translation.

Table 1 shows a part of a part of speech table for Korean and Japanese created by experiments.

[0044]

[Table 1]

For example, as shown in the first row of Table 1, a Korean common noun (Common Noun) is a common noun in Japanese.
(Common Noun) or Adverb. In addition,
Words are content words (nouns, verbs, adjectives, adverbs, etc.)
And functional words (words such as particles and auxiliary verbs), but the functional words are excluded from the part-of-speech table. The reason for this is that the correspondence of the function words largely depends on the words themselves, so that it is inappropriate to handle the function words at the part of speech level.

In the part of speech table, when the part of speech of the target word j matches one of the parts of speech corresponding to the part of speech of the source word k, POS_SCORE ( k, j) is assigned 1. Otherwise, POS_SCOR
Assign 0 as the value of E.

FIG. 2 shows the procedure of a word-to-word word correspondence process.

In the one-to-one word association process,
First, a word association process using a bilingual dictionary between the source language and the target language, that is, a word association process using WORD_SCORE is performed (step 11).

Next, for words that have not been associated by the processing in step 11, word association processing using a thesaurus for each of the two languages, that is,
A word association process using SEM_SCORE is performed (step 12).

Finally, step 11 and step 12
Is performed by using the part-of-speech table indicating the correspondence between parts of speech between the two languages, that is, the word association processing using POS_SCORE is performed on the word that has not been associated by the processing of step (step S1). 13).

Although the types of scores used in steps 11, 12 and 13 are different, the same algorithm is used for associating words, so the common algorithm will be described.

We assume that in languages closely related, such as Japanese and Korean, the basic word order is almost the same. It is assumed that a set of bilingual sentences subjected to morphological analysis is given. The bilingual sentence is a source language sentence (source sentence) consisting of m words K = <k ₁ , k ₂ , k ₃ ,.
_m > and a target language sentence (target sentence) J = <j ₁ , j ₂ , j ₃ ,... j _n > composed of _n words.

First, an m * n matrix c is created. The element c [x, y] has a subsequence <k ₁ , k ₂ , k ₃ ,... K _x ><j ₁ , as shown in the following equation (6).
j ₂ , j ₃ ,... j _y >.

[0054]

(Equation 3)

FIG. 3 shows that K = <A, B, Y, E> and J = <
It shows the value of each element c [x, y] of the 4 × 5 matrix when a, b, c, d, e>. Here, “a” is “A”, “b” is “B”, and “e” is “
E ".

Next, as shown in FIG. 4, a route having the maximum total score is searched. Then, as shown in FIG. 5, words are associated based on the searched route.

FIG. 6 shows an algorithm for associating the above words.

First, the length of the source sentence K (the number of words forming the source sentence K) is set in a variable m, and the length of the target sentence J (the number of words forming the target sentence J) is set in a variable n. Is set (step 101).

When the variable x is a value from 1 to m, the element c
Let the value of [x, 0] be 0. That is, first, x = 1 (step 102) and c [x, 0] = 0 (step 103). It is determined whether or not x is greater than or equal to m (step 104). If x is smaller than m, the value of x is incremented by 1 (x = x + 1) (step 10).
5), go to step 103; Thus, the processing of steps 103, 104 and 105 is repeatedly performed.

Then, in step 104, x is m
If it is determined that this is the case, the variable y is set to a value from 0 to n, and the value of the element c [0, y] is set to 0. That is, first, y = 0 (step 106), and c [0, y] = 0.
(Step 107). It is determined whether or not y is greater than or equal to n (step 108). If y is smaller than n, y is determined.
Is incremented by one (y = y + 1) (step 109), and the process proceeds to step 107. Thus, the processing of steps 107, 108, and 109 is repeatedly performed.

Then, in step 108, y becomes n
If it is determined that the above is satisfied, x = 1, 2,..., M, y =
For each combination of [x, y] in the range of 1, 2,..., N, a partial score c [x, y] and a type of arrow b [x, y] indicating a route are obtained. Perform processing.

That is, first, x = 1 (step 110), and y = 1 (step 111). Then, it is determined whether the first condition of SCORE (K [x], J [y])> 0 is satisfied (step 112). If the first condition is satisfied, the value of the element c [x, y] is changed to ｛c [x−
1, y−1] + SCORE (K [x], J [y])}, and the value of b [x, y] is set to “11” representing a diagonal arrow (step 113). ). And
Proceed to step 117.

If it is determined in step 112 that the first condition is not satisfied, c [x−1, y] ≧ c [x,
y-1] is determined (step 114). If the second condition is satisfied, the value of the element c [x, y] is set to {c [x-1, y]}, and the value of b [x, y] is set to the upward arrow (upper arro
w) is set to "01" (step 115). Then, the process proceeds to step 117.

In step 114, if the second condition is not satisfied, the value of the element c [x, y] is changed to {c
[x, y-1]}, and b [x, y] is set to "10" representing a left arrow (step 116). Then, the process proceeds to step 117.

At step 117, it is determined whether or not y is not smaller than n. If y is smaller than n, the value of y is incremented by 1 (y = y + 1) (step 11).
8) Go to step 112. Thus, the processing of steps 112 to 118 is repeatedly performed.

Then, in step 117, y becomes n
If it is determined that it is not less than x, it is determined whether or not x is not less than m (step 119).
Is incremented by 1 (x = x + 1) (step 120), and the process proceeds to step 111. And
Steps 111 to 120 are repeatedly performed.

In this way, if it is determined in step 119 that x is not less than m, the current processing is terminated. Then, the route having the maximum total score is output by following the matrix b from b [m, n]. b [x, y] is "11"
The word K [x], which is a (diagonal arrow), corresponds to the word J [y].

In the word matching process using WORD_SCORE in step 11 of FIG.
A word association process is performed. At this time, SCORE (K [x], J
WORD_SCORE is used as the value of [y]).

In the word associating process using SEM_SCORE in step 12 in FIG. 2, the above algorithm is applied to K and J from which the words associated by step 11 have been removed. At this time, SCORE (K
SEM_SCORE is used as the value of [x], J [y]).

In the word associating process using POS_SCORE in step 13 in FIG.
The above algorithm is applied to K and J from which the word associated by the process 2 has been removed.
At this time, POS_SCORE is used as the value of SCORE (K [x], J [y]).

[2-3] Description of processing for associating compound words

The correspondence of words in a bilingual sentence varies from correspondence for each word to correspondence for several words. Therefore, after the word-to-word one-to-one word associating process, the process of associating the compound words is performed.

FIG. 7 shows a procedure for associating compound words. In the synthetic word associating process, first, an associating process using a synthetic word dictionary between two languages is performed (step 21).

Next, an associating process is performed using a compound word part-of-speech table representing the correspondence between the parts of speech of the compound words between the two languages (step 22).

Table 2 shows a part of the bilingual compound word dictionary used in the processing of step 21.

[0076]

[Table 2]

Table 3 shows a part of the compound word part-of-speech table used in the processing of step 22.

[0078]

[Table 3]

Next, the algorithm of the associating process using the composite word dictionary between the two languages in step 21 will be described.

In the partially corresponded bilingual sentence, a word that is not associated is defined as a word to be processed in the compound word association process.

First, a processing target word is retrieved from a source sentence (Korean sentence) to extract one processing target word P. This search is performed from the left end of the source sentence. Next, in the Korean column of the compound word dictionary between two languages,
It is checked whether a word or a word string (hereinafter, referred to as Q) starting with the extracted processing target word P exists.

When Q exists in the Korean column of the composite word dictionary between two languages, a word string corresponding to Q (hereinafter, referred to as R) is read from the corresponding Japanese column in the composite word dictionary.
Is extracted. Then, it is checked whether or not Q starting with the processing target word P exists in the source sentence of the partially-translated bilingual sentence, and whether or not R corresponding to Q exists in the target sentence.

When Q exists in the source sentence and R exists in the target sentence, Q and R are associated with each other. In this case, at least one of Q and R is a compound word. Such processing is performed on each processing target word of the source sentence. If there is a correspondence between a plurality of sets of compound words partially overlapping, the longest compound word is adopted.

After performing a matching process based on a search for the processing target word P from the left end of the source sentence (a matching process using a compound word dictionary), a search for the processing target word P from the right end of the source sentence is performed. An associating process is performed based on the information.

In the above example, the processing target word P is searched from the source sentence (Korean sentence), one processing target word P is extracted, and a predetermined process is performed based on the extracted processing target word P. The processing target word P is searched from the target sentence (Japanese sentence), one processing target word P is extracted, and a predetermined process is performed based on the extracted processing target word P. Good.

Next, an explanation will be given of the algorithm of the association process using the compound word part-of-speech table representing the correspondence between the parts of speech of the compound words in both languages in step 22.

In the bilingual sentence partially corresponded by the processing up to the step 21, a word that is not matched is defined as a processing target word in the compound word matching processing.

First, a processing target word is retrieved from a source sentence (Korean sentence) to extract one processing target word P. This search is performed from the left end of the source sentence. Next, it is checked whether or not a part-of-speech sequence S starting with the part-of-speech of the extracted processing target word P exists in the Korean column of the compound word part-of-speech table.

In the Korean column of the compound word part-of-speech table,
If the part-of-speech sequence S exists, the part-of-speech sequence T corresponding to the part-of-speech sequence S is extracted from the corresponding Japanese column in the compound word part-of-speech table. Then, it is checked whether or not the part-of-speech sequence S starting with the processing target word P exists in the source sentence of the partially-translated bilingual sentence, and whether the part-of-speech sequence T exists in the target sentence.

If the part-of-speech sequence S exists in the source sentence and the part-of-speech sequence T exists in the target sentence, S and T are associated with each other. Such processing is performed on each processing target word of the source sentence. If there is a correspondence between a plurality of sets of compound words partially overlapping, the longest compound word is adopted.

After performing an associating process based on the search of the processing target word P from the left end of the source sentence (associating process using the compound word part-of-speech table), a search of the processing target word P from the right end of the source sentence is performed. Is performed based on.

In the above example, the processing target word P is searched from the source sentence (Korean sentence), one processing target word P is extracted, and a predetermined process is performed based on the extracted processing target word P. The processing target word P is searched from the target sentence (Japanese sentence), one processing target word P is extracted, and a predetermined process is performed based on the extracted processing target word P. Good.

[2-4] Description of associating process for uncorresponding words (step 5 in FIG. 1)

In the associating process for uncorresponding words (step 5 in FIG. 1), it is assumed that the word order of the source sentence and the target sentence of the bilingual sentence is substantially the same, and the word is not yet associated in the bilingual sentence. In the portions, those that correspond in position are associated with each other.

[3] Experiment In order to prove the feasibility of the present invention, a preliminary experiment was conducted between Korean and Japanese.

[3-1] Experimental Conditions [3-1-1] Vocabulary Resources in Two Languages Table 4 shows the statistics of the five resources we used in this experiment.

[0097]

[Table 4]

We utilized the resources of our previous translation project (see reference 11): (i) a bilingual dictionary, (iv) a bilingual dictionary, and (ii) a thesaurus. (i) and (iv) were generalized to the part-of-speech level and checked manually to obtain (iii) and (v).

Reference 11: Sumita, E., Yamada, S., Yama
moto, K., Paul, M., Kashioka, H., Ishikawa, K., an
d Shirai, S. (1999) Solutions to Problems Inherent
in Spoken-language Translation: The ATR-MATRIX App
roach, Proc. of 7th MT Summit, pp.229-235.

[3-1-2] Translated sentences We have a Korean and Japanese phrasebook for foreign travelers (a collection of typical Korean sentences and their Japanese translations). Was used as a bilingual sentence. Bilingual statistics are summarized in Table 5. The length of the sentence is relatively short, about 8.1 words in Korean on average.

[0101]

[Table 5]

[3-2] Accuracy We randomly selected 10 sentence spares from the above translated text.
Zero sets were selected and associated by the method described in the above embodiment. A bilingual speaker evaluated the results for each response. Table 6 shows the evaluation results. 695 correspondences were obtained for 100 sentence spares. The correct association was 594.

[0103]

[Table 6]

We did not distinguish between the following two unhandled cases: The two cases are: (1) Inaccurate: the association should have been made, but the system could not achieve this, (2) Exact: there was no corresponding in sentence spare.

If all unsupported cases are regarded as incorrect,
The recall will be calculated lower. The recall in this case is defined by the following equation (7).

[0106]

(Equation 4)

In our experiments, as shown in Table 7, high recall quasi_recall (93.7%) and high precision prec
ision (91.2%) was achieved.

[0108]

[Table 7]

Table 8 shows each step (step 3 in FIG. 1 (steps 11, 12, and 13 in FIG. 2),
4 (steps 21 and 22 in FIG. 7) and step 5) in FIG.

[0110]

[Table 8]

The word matching process using WORD_SCORE (step 11 in FIG. 2) is the most effective and accurate step. The remaining steps, although degraded in accuracy, contribute significantly to recall.

[0112]

According to the present invention, a method of associating a bilingual sentence with a high recall and high accuracy can be obtained.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an overall processing procedure of a method for associating a bilingual sentence with a word.

FIG. 2 is a flowchart illustrating a procedure of a word-to-word word associating process;

FIG. 3 shows K = <A, B, Y, E>, J = <a, b, c,
It is a schematic diagram which shows the example of the value of the partial score c [x, y] of the 4x5 matrix in case d, e>.

FIG. 4 is a schematic diagram showing a route having a maximum total score in FIG. 3;

FIG. 5 is a schematic diagram showing words associated with each other based on FIG. 4;

FIG. 6 is a flowchart showing an algorithm used for a word-to-word word correspondence process.

FIG. 7 is a flowchart illustrating a synthetic word association processing procedure;

Claims (14)

[Claims] 1. A method for associating a bilingual sentence comprising a source language sentence and a target language sentence corresponding thereto with a word-to-word correspondence between the source language sentence and the target language sentence, A first step of calculating a first word correspondence degree using a bilingual dictionary between the two languages, and a second step of obtaining a word correspondence combination that maximizes the sum of the first word correspondence degrees in a bilingual sentence A method for associating a bilingual sentence with words, comprising: 2. The first degree of word correspondence between an arbitrary source language word and an arbitrary target language word is as follows: the target language word is a translation of the source language word in a bilingual dictionary. 2. The method for associating a bilingual sentence with a word according to claim 1, wherein the word is determined based on whether or not the word matches one of the words. 3. For a part of the entire bilingual sentence that has not yet been associated in the second step, a synonym dictionary for each of the two languages is created for each one-to-one word correspondence between the two languages. Utilizing the third step of calculating the second word correspondence, and the sum of the second word correspondence is maximum in a portion of the entire bilingual sentence that has not yet been matched by the second step. 4. The method according to claim 1, further comprising: a fourth step of obtaining a combination of word correspondences. 4. A second word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is determined by a set of semantic classes to which the word in the source language belongs obtained from a thesaurus and the target language 4. The method according to claim 3, wherein the calculation is performed based on a set of semantic classes to which the word belongs. 5. A second word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is determined by a set of semantic classes to which the word in the source language belongs obtained from a thesaurus and the target language 5. The method according to claim 4, wherein the ratio is represented by a ratio of the size of the common set to the size of the union of the set of the semantic classes to which the word belongs. 6. A part of the entire bilingual sentence that has not yet been associated in the second and fourth steps, for each one-to-one word correspondence between the two languages. The fifth step of calculating the third degree of word correspondence using the part-of-speech table representing the part-of-speech correspondence, and the parts of the entire bilingual sentence that have not yet been matched by the second and fourth steps. 6. A bilingual sentence word according to any one of claims 3, 4 and 5, further comprising: a sixth step of obtaining a combination of word correspondences that maximizes the third sum of word correspondence degrees. Mapping method. 7. The third word correspondence between a word in an arbitrary source language and a word in an arbitrary target language is as follows: the part of speech of the word in the target language is the part of speech of the word in the source language in the part of speech table. 7. It is determined based on whether or not it matches any of the parts of speech corresponding to.
The word correspondence method of the bilingual sentence described in. 8. The second step, the fourth step and the sixth step.
In the bilingual sentence partially supported in the step,
8. The method according to claim 6, further comprising a seventh step of associating one word to a composite word or a composite word to a composite word using a composite word dictionary between two languages. How to match words in the bilingual sentence. 9. A bilingual sentence partially corresponding in the second, fourth, sixth, and seventh steps, using a compound word part-of-speech table representing the correspondence of part of speech of a compound word between two languages. 9. The method for associating a bilingual sentence with a word according to claim 8, further comprising an eighth step of associating one word-to-synthesis word or a synthetic word-to-synthesis word. 10. Assuming that the word order of the source language sentence and the target language sentence forming the bilingual sentence is substantially the same, the second step, the fourth step, the sixth step, the seventh step, and The method according to claim 9, further comprising a ninth step of associating a positionally corresponding part in a part that has not yet been associated in the eighth step. 11. A bilingual sentence partially corresponding in the second step, using a compound word dictionary between two languages,
3. The method for associating a bilingual sentence with a word according to claim 1, further comprising a tenth step of associating one word to a compound word or a compound word to a compound word. 12. Assuming that the word order of the source language sentence and the target language sentence constituting the bilingual sentence is almost the same, no association has been made yet in the second and tenth steps of the entire bilingual sentence. 12. The method for associating words in a bilingual sentence according to claim 11, further comprising an eleventh step of associating parts corresponding in position with each other. 13. A bilingual sentence partially corresponded in the second step and the tenth step, using a compound word part-of-speech table indicating the correspondence of part of speech of a compound word between two languages, one word pair compound word or compound word The method according to claim 11, further comprising a twelfth step of associating a word pair with a composite word. 14. Assuming that the source language sentence constituting the bilingual sentence and the target language sentence have substantially the same word order, the second, tenth, and twelfth steps of the entire bilingual sentence are still in correspondence. 14. The method for associating a bilingual sentence with a word according to claim 13, further comprising a thirteenth step of associating a positionally corresponding part in a part that has not been performed.